a2a cloud
Start building
a2a cloud vs OpenAI AgentKit

a2a cloud vs OpenAI AgentKit for deploying AI agents

OpenAI's AgentKit and Agents SDK are a genuinely good way to build an agent — clean primitives for tools, handoffs, and guardrails on OpenAI's models. But building the agent and hosting it in production are different jobs. a2a cloud is the framework- and model-agnostic runtime: it deploys the agent, gives it a database and surfaces, and proves what it did.

Deploy an agentView architectureWhy not DIY?
kindlock-indatabaseagent surfacesauthorityproof
honest take

Where OpenAI AgentKit is genuinely strong — and where a2a cloud is different.

OpenAI AgentKit is good at

A clean way to build on OpenAI

AgentKit and the Agents SDK give you well-designed building blocks — tool calls, agent handoffs, guardrails, and tracing — tightly integrated with OpenAI's models. If your stack is OpenAI-first and you want an ergonomic authoring experience, it's an excellent place to build.

a2a cloud adds

Model-agnostic hosting with proof

a2a cloud hosts agents with managed Postgres, MCP, an OpenAPI gateway, a frontend, and auth. Signed grants cover grant-backed operations. The runtime signs Agent API calls; the gateway signs public /invoke and MCP tools/call.

side-by-side

a2a cloud vs OpenAI AgentKit, dimension by dimension.

A fair comparison. Both columns are accurate as we understand the products today — the difference is what the runtime owns by default.

dimension
OpenAI AgentKit
a2a cloud
kind
A framework and toolkit for building agents on OpenAI's models and platform.
A deployment platform that hosts the agent — runtime, Postgres, MCP, API, and frontend.
lock-in
Oriented around OpenAI models, tools, and the OpenAI platform surface.
Framework- and model-agnostic — deploy Python or TypeScript agents on any framework or LLM.
database
Bring your own persistence — the SDK builds the agent, not a managed database.
A managed Postgres database per agent, provisioned automatically on deploy.
agent surfaces
Tools and handoffs in code; you still host the MCP server, API gateway, and auth.
MCP server, OpenAPI gateway, and auth ship with the deploy. Every agent tool is also an MCP tool.
authority
API keys and tool credentials — broad, long-lived secrets the agent holds.
Ed25519-signed grants scope grant-backed delegation and workspace operations; user-configured secrets remain container environment variables.
proof
Traces and platform logs — useful for debugging, not per-run signed evidence.
The runtime signs and stores receipts for Agent API calls, and signs public /invoke and MCP tools/call responses as they go out.
how to choose

Pick the tool that matches the job.

Reach for OpenAI AgentKit when

  • Your stack is OpenAI-first and you want its native authoring tools.
  • You're building the agent's logic and will host it yourself.
  • Tight integration with OpenAI models and traces is the priority.

Reach for a2a cloud when

  • You want to deploy the agent — DB, MCP, API, frontend — not just build it.
  • You need framework- and model-agnostic hosting with no vendor lock-in.
  • You need scoped grants and signed receipts for governance and proof.
don't trust the agent

Trust the receipt.

a2a cloud deploys any agent — LangGraph, OpenAI Agents SDK, CrewAI, or custom — and ships it with a managed Postgres database, an MCP server, an API, a frontend, and an Ed25519-signed receipts on governed execution paths. Scoped grants, no ambient production access. One deploy, the whole agent app, with proof.

Deploy from the CLIDescribe an agent insteadPlan a pilot